docs/drivers/AbCip-Operability.md

@@ -150,3 +150,150 @@ rather than a separate tier of scan-class definitions.
   with per-tag scan rate when a slow bucket starves a fast one.
 - S7 driver `ScanGroup` model in `src/.../S7DriverOptions.cs` — the
   named-group form of the same idea.
+
+## Write deadband / write-on-change
+
+PR abcip-4.2 ships the second operability knob: per-tag write coalescing,
+the *write-side* companion to the read-side deadband already shipped at the
+OPC UA monitored-item layer. The driver remembers the value last
+successfully written for a tag and can suppress redundant or below-threshold
+follow-up writes — they return `Good` to the OPC UA client without ever
+hitting the wire.
+
+### What it is
+
+- **`AbCipTagDefinition.WriteDeadband`** (`double?`, default `null`) —
+  numeric absolute-difference threshold. When set, a write whose
+  `|new − last|` is below the deadband is suppressed.
+- **`AbCipTagDefinition.WriteOnChange`** (`bool`, default `false`) —
+  equality gate. When set, a write whose value equals the last successfully
+  written value is suppressed.
+
+Both knobs combine on the same tag. For numerics, the deadband path takes
+priority; the equality fallback covers the cases the deadband doesn't (BOOL
+setpoints, STRING constants, `WriteDeadband=0`, etc).
+
+### Worked setpoint-jitter example
+
+A motor speed setpoint published from an HMI tends to wobble by a few
+ticks even when the operator hasn't touched it — UI rounding, Modbus
+gateway re-encoding, RPN script noise. With `WriteDeadband: 0.5`:
+
+```json
+{
+  "Tags": [
+    {
+      "Name": "Motor1.Speed.SP",
+      "DeviceHostAddress": "ab://10.0.0.5/1,0",
+      "TagPath": "Motor1.Speed.SP",
+      "DataType": "Real",
+      "WriteDeadband": 0.5
+    }
+  ]
+}
+```
+
+Sequence of writes from the HMI (one every 100 ms, no operator input):
+
+| Time | Value | `\|new − last\|` | Wire? |
+|---|---|---|---|
+| 0 ms | 50.0 | n/a (first) | yes |
+| 100 ms | 50.2 | 0.2 < 0.5 | suppressed |
+| 200 ms | 50.3 | 0.3 < 0.5 | suppressed |
+| 300 ms | 50.6 | 0.6 ≥ 0.5 | yes |
+| 400 ms | 50.6 | 0.0 < 0.5 | suppressed |
+| 500 ms | 51.5 | 0.9 ≥ 0.5 | yes |
+
+Three writes hit the wire; three are suppressed. The OPC UA client sees
+`Good` on every call. The PLC sees only the values that actually crossed
+the deadband.
+
+### Combining with WriteOnChange
+
+A digital reset bit driven by a UI that pulses it at every cycle:
+
+```json
+{
+  "Name": "Conveyor.Reset",
+  "DeviceHostAddress": "ab://10.0.0.5/1,0",
+  "TagPath": "Conveyor.Reset",
+  "DataType": "Bool",
+  "WriteOnChange": true
+}
+```
+
+Three consecutive `false → false → false` writes from the UI collapse to
+one wire write (`false`, the first). When the operator clicks the reset
+button (`true`), that write passes; subsequent `true → true` repeats
+suppress until the UI clears it back to `false`.
+
+Numeric tags can also opt into both: `WriteDeadband: 0.5` plus
+`WriteOnChange: true` is well-defined — the deadband suppresses jitter, the
+equality gate suppresses exact repeats (which the deadband path also catches
+because `|0| < 0.5`, but having both set documents the operator's intent).
+
+### Special cases
+
+- **First write** always passes through. The coalescer has no prior value
+  to compare against, so the first write of any tag pays the full
+  round-trip and seeds the cache.
+- **NaN / Infinity** bypass deadband suppression. IEEE-754 comparisons
+  against NaN are undefined and a stale `+Inf` shouldn't silently swallow
+  a real reset; the wire decides. `WriteOnChange` equality on NaN still
+  follows .NET semantics (`Equals(NaN, NaN) == true` for `double` boxed in
+  `object`), so a `WriteOnChange` tag stuck on NaN will suppress repeats
+  until something else writes a real value.
+- **Failed writes** do *not* seed the cache. If the wire write fails, the
+  next attempt with the same value still hits the wire because the
+  coalescer never recorded a "last successful value" for it.
+- **Reconnect drops the cache**. The driver's host-state probe transitions
+  `Stopped → Running` after a reconnect; both transitions reset the
+  per-device coalescer cache, so the first post-reconnect write of any
+  value pays the full round-trip. The PLC may have been restarted while
+  the driver was offline and our cached "we already wrote 42" is stale.
+- **Two devices, same tag address**. The cache is keyed on
+  `(deviceHostAddress, tagAddress)` so two PLCs running the same Logix
+  program keep independent caches — writing 42 to A doesn't suppress
+  writing 42 to B.
+- **Bit-in-DINT writes** consult the coalescer too, so a UI that pulses
+  `Flags.3` at every cycle benefits from the same `WriteOnChange`
+  suppression as a plain BOOL tag.
+- **Plain back-compat tags** (no `WriteDeadband`, no `WriteOnChange`)
+  take a fast-path through the coalescer that increments only the
+  `WritesPassedThrough` counter — no dictionary lookup, no allocation. The
+  knobs are zero-overhead opt-in.
+
+### Diagnostics
+
+The driver surfaces two counters through `DriverHealth.Diagnostics` (the
+same path the `driver-diagnostics` RPC + Admin UI render for Modbus / S7 /
+OPC UA Client):
+
+- `AbCip.WritesSuppressed` — total writes the coalescer skipped.
+- `AbCip.WritesPassedThrough` — total writes that hit the wire after
+  consulting the coalescer.
+
+Their ratio is the "wire savings" headline. A deployment with `0`
+suppressions either has no tags opted in or has the deadband too tight /
+the equality threshold too loose; revisit the per-tag config.
+
+### Verification
+
+- **Unit**: `AbCipWriteDeadbandTests` (`tests/.../AbCip.Tests`). Asserts
+  the deadband math, the equality fallback, the first-write pass-through,
+  reset-on-reconnect, two-device cache independence, suppressed-Good
+  status, NaN bypass, the back-compat fast path, and DTO round-trip.
+- **Integration**: `AbCipWriteDeadbandTests`
+  (`tests/.../AbCip.IntegrationTests`). Drives a 5-write jittery sequence
+  with `WriteDeadband: 1.0` against a live `ab_server` and asserts the
+  driver's diagnostics counter matches the expected suppression count.
+- **E2E**: `scripts/e2e/test-abcip.ps1` — see the *WriteCoalesce*
+  assertion.
+
+### Cross-references
+
+- `docs/drivers/AbServer-Test-Fixture.md` §7 — capability surfaces beyond
+  read; mentions write-coalesce coverage.
+- Modbus driver — read-side deadband in `ModbusDriver` predates this
+  write-side equivalent; the config shape is intentionally similar.
+- Kepware "Deadband (write)" knob — this is the AB CIP equivalent.

docs/drivers/AbServer-Test-Fixture.md

@@ -139,14 +139,19 @@ the RMW path is not exercised end-to-end.
 
 No smoke test for:
 
-- `IWritable.WriteAsync`
+- `IWritable.WriteAsync` — atomic write coverage; PR abcip-4.2 added a
+  multi-write *suppression* smoke (jittery 5-write sequence with
+  `WriteDeadband: 1.0` against `ab_server`, asserting the driver's
+  diagnostics counter matches the expected suppression count) but pure
+  atomic-write coverage end-to-end is still unit-only.
 - `ITagDiscovery.DiscoverAsync` (`@tags` walker)
 - `ISubscribable.SubscribeAsync` (poll-group engine)
 - `IHostConnectivityProbe` state transitions under wire failure
 - `IPerCallHostResolver` multi-device routing
 
 The driver implements all of these + they have unit coverage, but the only
-end-to-end path `ab_server` validates today is atomic `ReadAsync`.
+end-to-end paths `ab_server` validates today are atomic `ReadAsync` and
+write-deadband / write-on-change suppression.
 
 ## Logix Emulate golden-box tier
 

scripts/e2e/test-abcip.ps1

@@ -182,5 +182,31 @@ if ($FastBridgeNodeId -and $SlowBridgeNodeId) {
     $results += [PSCustomObject]@{ Name = "PerTagScanRate"; Passed = $passed; Detail = $detail }
 }
 
+# PR abcip-4.2 — write-coalesce assertion. Writes the same value twice through the OPC UA
+# server and verifies the PLC-side state reflects only one wire write. The driver-side
+# diagnostics counter (AbCip.WritesSuppressed) is the authoritative signal, but ab_server
+# itself doesn't expose a "writes received" counter so this script-level check is intentionally
+# observational — it primes the tag with a baseline, writes the same value twice, and reads
+# back to confirm the value matches without surfacing additional state changes. The unit + integration
+# tests do the strict "exactly N suppressions" math; this is the e2e shape proof.
+$coalesceValue = Get-Random -Minimum 60000 -Maximum 69999
+Write-Header "WriteCoalesce (baseline=$coalesceValue, two redundant writes)"
+$writeArgs = @("write") + $commonAbCip + @("-t", $TagPath, "--type", "DInt", "-v", $coalesceValue)
+& $abcipCli.Exe @($abcipCli.Args + $writeArgs) | Out-Null
+& $abcipCli.Exe @($abcipCli.Args + $writeArgs) | Out-Null
+& $abcipCli.Exe @($abcipCli.Args + $writeArgs) | Out-Null
+$readArgs = @("read") + $commonAbCip + @("-t", $TagPath, "--type", "DInt")
+$readOut = & $abcipCli.Exe @($abcipCli.Args + $readArgs)
+$coalesceMatch = ($readOut -join "`n") -match "$coalesceValue"
+$results += [PSCustomObject]@{
+    Name   = "WriteCoalesce"
+    Passed = $coalesceMatch
+    Detail = if ($coalesceMatch) {
+        "three identical writes of $coalesceValue produced the expected readback (driver-side WritesSuppressed counter exposed via driver-diagnostics RPC)"
+    } else {
+        "three identical writes did not converge on $coalesceValue — got '$readOut'"
+    }
+}
+
 Write-Summary -Title "AB CIP e2e" -Results $results
 if ($results | Where-Object { -not $_.Passed }) { exit 1 }

src/ZB.MOM.WW.OtOpcUa.Driver.AbCip/AbCipDriver.cs

@@ -35,6 +35,7 @@ public sealed class AbCipDriver : IDriver, IReadable, IWritable, ITagDiscovery,
     private readonly Dictionary<string, AbCipTagDefinition> _tagsByName = new(StringComparer.OrdinalIgnoreCase);
     private readonly AbCipAlarmProjection _alarmProjection;
     private readonly SemaphoreSlim _discoverySemaphore = new(1, 1);
+    private readonly AbCipWriteCoalescer _writeCoalescer = new();
     private DriverHealth _health = new(DriverState.Unknown, null, null);
 
     public event EventHandler<DataChangeEventArgs>? OnDataChange;
@@ -415,6 +416,10 @@ public sealed class AbCipDriver : IDriver, IReadable, IWritable, ITagDiscovery,
         }
         _devices.Clear();
         _tagsByName.Clear();
+        // PR abcip-4.2 — wipe the write-coalescer cache on shutdown. Reinitializing the driver
+        // (Tier-B remediation) starts from a clean slate so the first write after restart pays
+        // the full round-trip rather than reusing stale cached state.
+        _writeCoalescer.ResetAll();
         _health = new DriverHealth(DriverState.Unknown, _health.LastSuccessfulRead, null);
     }
 
@@ -637,6 +642,13 @@ public sealed class AbCipDriver : IDriver, IReadable, IWritable, ITagDiscovery,
             state.HostState = newState;
             state.HostStateChangedUtc = DateTime.UtcNow;
         }
+        // PR abcip-4.2 — drop the per-device write-coalescer cache when we lose the wire. The
+        // PLC may have been restarted while we were offline + our cached "we already wrote 42"
+        // is no longer valid PLC state. Reset on the Stopped transition (and again on the
+        // recovery edge for safety) so the first post-reconnect write of any value pays the
+        // full round-trip + the coalescer rebuilds its cache from the new baseline.
+        if (newState == HostState.Stopped || newState == HostState.Running)
+            _writeCoalescer.Reset(state.Options.HostAddress);
         OnHostStatusChanged?.Invoke(this,
             new HostStatusChangedEventArgs(state.Options.HostAddress, old, newState));
     }
@@ -1255,6 +1267,16 @@ public sealed class AbCipDriver : IDriver, IReadable, IWritable, ITagDiscovery,
         var def = entry.Definition;
         var w = entry.Request;
         var now = DateTime.UtcNow;
+
+        // PR abcip-4.2 — write deadband / write-on-change. Consult the coalescer first; a
+        // suppression decision returns Good without hitting libplctag so the OPC UA client sees
+        // the same write semantics it always has, the wire just doesn't move. Driver health is
+        // intentionally left alone on suppression — a coalesced write is neither a success nor
+        // a failure of the underlying connection. Bit-RMW writes go through their own path
+        // (ExecuteBitRmwWriteAsync) which has its own coalescer call site.
+        if (_writeCoalescer.ShouldSuppress(def.DeviceHostAddress, def, w.Value))
+            return (entry.OriginalIndex, AbCipStatusMapper.Good);
+
         try
         {
             var runtime = await EnsureTagRuntimeAsync(device, def, ct).ConfigureAwait(false);
@@ -1265,6 +1287,7 @@ public sealed class AbCipDriver : IDriver, IReadable, IWritable, ITagDiscovery,
             if (status == 0)
             {
                 _health = new DriverHealth(DriverState.Healthy, now, null);
+                _writeCoalescer.Record(def.DeviceHostAddress, def, w.Value);
                 return (entry.OriginalIndex, AbCipStatusMapper.Good);
             }
             return (entry.OriginalIndex, AbCipStatusMapper.MapLibplctagStatus(status));
@@ -1309,13 +1332,24 @@ public sealed class AbCipDriver : IDriver, IReadable, IWritable, ITagDiscovery,
     private async Task<uint> ExecuteBitRmwWriteAsync(
         DeviceState device, AbCipMultiWritePlanner.ClassifiedWrite entry, CancellationToken ct)
     {
+        // PR abcip-4.2 — bit-RMW writes go through the coalescer too. The deadband path is
+        // never useful on a single-bit BOOL (deadband < 1 collapses to equality) but
+        // WriteOnChange is — a UI that toggles a SetPoint.Reset bit at every cycle benefits
+        // from suppressing the redundant pulses.
+        var def = entry.Definition;
+        if (_writeCoalescer.ShouldSuppress(def.DeviceHostAddress, def, entry.Request.Value))
+            return AbCipStatusMapper.Good;
+
         try
         {
             var bit = entry.ParsedPath!.BitIndex!.Value;
             var code = await WriteBitInDIntAsync(device, entry.ParsedPath, bit, entry.Request.Value, ct)
                 .ConfigureAwait(false);
             if (code == AbCipStatusMapper.Good)
+            {
                 _health = new DriverHealth(DriverState.Healthy, DateTime.UtcNow, null);
+                _writeCoalescer.Record(def.DeviceHostAddress, def, entry.Request.Value);
+            }
             return code;
         }
         catch (OperationCanceledException)
@@ -1478,7 +1512,30 @@ public sealed class AbCipDriver : IDriver, IReadable, IWritable, ITagDiscovery,
         return runtime;
     }
 
-    public DriverHealth GetHealth() => _health;
+    public DriverHealth GetHealth() => _health with { Diagnostics = BuildDiagnostics() };
+
+    /// <summary>
+    ///     PR abcip-4.2 — driver-attributable counter snapshot exposed via
+    ///     <see cref="DriverHealth.Diagnostics"/> + the <c>driver-diagnostics</c> RPC. Names use
+    ///     the <c>"&lt;DriverType&gt;.&lt;Counter&gt;"</c> convention so the Admin UI can render
+    ///     them alongside Modbus / S7 / OPC UA Client metrics without per-driver special-casing.
+    ///     Counters today: <c>AbCip.WritesSuppressed</c> (writes the coalescer skipped because
+    ///     deadband / write-on-change suppressed them) and <c>AbCip.WritesPassedThrough</c>
+    ///     (writes that hit the wire after consulting the coalescer). Future PRs add CIP-level
+    ///     counters (Forward Open count, multi-service-packet ratio, etc.) by extending this
+    ///     dictionary.
+    /// </summary>
+    private IReadOnlyDictionary<string, double> BuildDiagnostics() => new Dictionary<string, double>
+    {
+        ["AbCip.WritesSuppressed"] = _writeCoalescer.TotalWritesSuppressed,
+        ["AbCip.WritesPassedThrough"] = _writeCoalescer.TotalWritesPassedThrough,
+    };
+
+    /// <summary>
+    ///     Test seam — exposes the live coalescer for unit tests that want to inspect counters
+    ///     without rebuilding the diagnostics dictionary on every assertion.
+    /// </summary>
+    internal AbCipWriteCoalescer WriteCoalescer => _writeCoalescer;
 
     /// <summary>
     ///     CLR-visible allocation footprint only — libplctag's native heap is invisible to the

src/ZB.MOM.WW.OtOpcUa.Driver.AbCip/AbCipDriverFactoryExtensions.cs

@@ -86,7 +86,11 @@ public static class AbCipDriverFactoryExtensions
                 : null,
             SafetyTag: t.SafetyTag ?? false,
             // PR abcip-4.1 — per-tag scan rate override; null means "use subscription default".
-            ScanRateMs: t.ScanRateMs);
+            ScanRateMs: t.ScanRateMs,
+            // PR abcip-4.2 — per-tag write-deadband + write-on-change. Both default to "off"
+            // when absent so back-compat deployments behave exactly as before.
+            WriteDeadband: t.WriteDeadband,
+            WriteOnChange: t.WriteOnChange ?? false);
 
     private static T ParseEnum<T>(string? raw, string? tagName, string driverInstanceId, string field,
         T? fallback = null) where T : struct, Enum
@@ -180,6 +184,22 @@ public static class AbCipDriverFactoryExtensions
         ///     that don't set the knob. Mirrors Kepware's "scan classes" model.
         /// </summary>
         public int? ScanRateMs { get; init; }
+
+        /// <summary>
+        ///     PR abcip-4.2 — optional numeric write deadband. When set, the driver skips a
+        ///     wire write whose absolute difference from the previous successfully-written
+        ///     value falls below this threshold. Suppressed writes still return <c>Good</c>.
+        ///     <c>null</c> = no numeric suppression (back-compat default).
+        /// </summary>
+        public double? WriteDeadband { get; init; }
+
+        /// <summary>
+        ///     PR abcip-4.2 — optional write-on-change gate. When <c>true</c>, the driver
+        ///     skips a wire write whose value equals the previous successfully-written value.
+        ///     Combines with <see cref="WriteDeadband"/> on numeric tags (deadband path takes
+        ///     priority for numerics). Default <c>false</c> — every write reaches the wire.
+        /// </summary>
+        public bool? WriteOnChange { get; init; }
     }
 
     internal sealed class AbCipMemberDto

src/ZB.MOM.WW.OtOpcUa.Driver.AbCip/AbCipDriverOptions.cs

@@ -279,6 +279,20 @@ public enum AddressingMode
 ///     <c>ScanRateMs &lt; 100</c> is clamped up. UDT member tags inherit the parent tag's
 ///     <c>ScanRateMs</c> at member-fan-out time. See
 ///     <c>docs/drivers/AbCip-Operability.md</c> §"Per-tag scan rate".</param>
+/// <param name="WriteDeadband">PR abcip-4.2 — optional numeric write deadband. When set and both
+///     the previous successfully-written value and the new write are numeric, the driver suppresses
+///     the next write if <c>|new - last| &lt; WriteDeadband</c>. Suppressed writes still return
+///     <c>Good</c> so the OPC UA write semantics observed by clients are unchanged — the driver
+///     simply skips the wire round-trip. Mirrors Kepware's "Deadband (write)" knob and is the
+///     write-side companion to the read-side deadband already shipped at the OPC UA monitored-item
+///     layer. NaN / Infinity values bypass suppression (let the wire decide). See
+///     <c>docs/drivers/AbCip-Operability.md</c> §"Write deadband / write-on-change".</param>
+/// <param name="WriteOnChange">PR abcip-4.2 — optional write-on-change gate. When <c>true</c> and
+///     the new write equals the previous successfully-written value, the driver suppresses the
+///     write (returns <c>Good</c> without hitting the wire). Combines with <see cref="WriteDeadband"/>
+///     for numeric tags — the deadband test takes priority for numerics, equality is the fallback
+///     for non-numeric types (BOOL setpoints, STRING constants, etc.). Default <c>false</c> —
+///     legacy behaviour where every write goes to the wire.</param>
 public sealed record AbCipTagDefinition(
     string Name,
     string DeviceHostAddress,
@@ -290,7 +304,9 @@ public sealed record AbCipTagDefinition(
     bool SafetyTag = false,
     int? StringLength = null,
     string? Description = null,
-    int? ScanRateMs = null);
+    int? ScanRateMs = null,
+    double? WriteDeadband = null,
+    bool WriteOnChange = false);
 
 /// <summary>
 ///     One declared member of a UDT tag. Name is the member identifier on the PLC (e.g. <c>Speed</c>,

src/ZB.MOM.WW.OtOpcUa.Driver.AbCip/AbCipWriteCoalescer.cs

@@ -0,0 +1,211 @@
+using System.Collections.Concurrent;
+
+namespace ZB.MOM.WW.OtOpcUa.Driver.AbCip;
+
+/// <summary>
+///     PR abcip-4.2 — per-tag last-successfully-written-value cache supporting
+///     <see cref="AbCipTagDefinition.WriteDeadband"/> + <see cref="AbCipTagDefinition.WriteOnChange"/>
+///     suppression in <see cref="AbCipDriver.WriteAsync"/>. Keys are
+///     <c>(deviceHostAddress, tagAddress)</c>: the same Logix tag served from two devices
+///     keeps independent caches because the underlying PLC state is independent. Counters
+///     (<see cref="TotalWritesSuppressed"/>, <see cref="TotalWritesPassedThrough"/>) feed
+///     <c>AbCip.WritesSuppressed</c> / <c>AbCip.WritesPassedThrough</c> in the driver
+///     diagnostics surface.
+/// </summary>
+/// <remarks>
+///     <para>The coalescer is consulted *before* the wire write; only successful writes call
+///     <see cref="Record"/> so a failed write does not poison the cache (next attempt with the
+///     same value still hits the wire because no last-value was ever recorded for it).
+///     <see cref="Reset"/> wipes the per-device entries on reconnect / shutdown — the PLC may
+///     have been restarted and our cached "we already wrote 42" is no longer valid PLC state.</para>
+///
+///     <para>Suppression rules:</para>
+///     <list type="bullet">
+///         <item>No prior recorded value → not suppressed (first write always passes through).</item>
+///         <item><see cref="AbCipTagDefinition.WriteDeadband"/> + both values numeric →
+///             <c>|new - last| &lt; deadband</c> suppresses. NaN / Infinity in either side bypass
+///             suppression; the wire decides.</item>
+///         <item><see cref="AbCipTagDefinition.WriteOnChange"/> set →
+///             <see cref="object.Equals(object?, object?)"/> equality suppresses. For numeric tags
+///             with a deadband configured, this still applies as the equality fallback when the
+///             deadband path doesn't trigger (e.g. exact equality with a 0 deadband).</item>
+///         <item>Neither knob set → never suppress (back-compat default).</item>
+///     </list>
+/// </remarks>
+internal sealed class AbCipWriteCoalescer
+{
+    private readonly ConcurrentDictionary<(string Device, string Tag), object?> _lastValues =
+        new(LastKeyComparer.Instance);
+
+    private long _totalWritesSuppressed;
+    private long _totalWritesPassedThrough;
+
+    /// <summary>Diagnostics counter — number of writes the coalescer told the driver to skip.</summary>
+    public long TotalWritesSuppressed => Interlocked.Read(ref _totalWritesSuppressed);
+
+    /// <summary>Diagnostics counter — number of writes that hit the wire after consulting the coalescer.</summary>
+    public long TotalWritesPassedThrough => Interlocked.Read(ref _totalWritesPassedThrough);
+
+    /// <summary>
+    ///     Decide whether <paramref name="newValue"/> should suppress the wire write for
+    ///     <paramref name="tag"/> on <paramref name="deviceHostAddress"/>. Increments the
+    ///     internal <see cref="TotalWritesSuppressed"/> / <see cref="TotalWritesPassedThrough"/>
+    ///     counter as a side effect so callers don't have to maintain a parallel tally.
+    /// </summary>
+    /// <returns>
+    ///     <c>true</c> when the write can be skipped (last value recorded + suppression rule
+    ///     fired). <c>false</c> when the write must hit the wire (no prior value, no rule
+    ///     active, or values differ enough).
+    /// </returns>
+    public bool ShouldSuppress(string deviceHostAddress, AbCipTagDefinition tag, object? newValue)
+    {
+        ArgumentNullException.ThrowIfNull(deviceHostAddress);
+        ArgumentNullException.ThrowIfNull(tag);
+
+        // Fast path — neither knob active. Skip the dictionary lookup entirely; this is the
+        // overwhelming common case in deployments that don't opt in.
+        if (!tag.WriteOnChange && !tag.WriteDeadband.HasValue)
+        {
+            Interlocked.Increment(ref _totalWritesPassedThrough);
+            return false;
+        }
+
+        var key = (deviceHostAddress, tag.TagPath);
+        if (!_lastValues.TryGetValue(key, out var lastValue))
+        {
+            // No prior recorded write — first write must always pass through so the PLC sees a
+            // baseline. The Record call after a successful write seeds the cache from this point.
+            Interlocked.Increment(ref _totalWritesPassedThrough);
+            return false;
+        }
+
+        if (TrySuppress(tag, lastValue, newValue))
+        {
+            Interlocked.Increment(ref _totalWritesSuppressed);
+            return true;
+        }
+
+        Interlocked.Increment(ref _totalWritesPassedThrough);
+        return false;
+    }
+
+    /// <summary>
+    ///     Record the value just successfully written so the next call to
+    ///     <see cref="ShouldSuppress"/> can compare against it. Called only from the
+    ///     <see cref="AbCipDriver"/> success branch — failed writes do not seed the cache.
+    /// </summary>
+    public void Record(string deviceHostAddress, AbCipTagDefinition tag, object? writtenValue)
+    {
+        ArgumentNullException.ThrowIfNull(deviceHostAddress);
+        ArgumentNullException.ThrowIfNull(tag);
+
+        // Only care about tags that opted in to either knob — pure-passthrough tags don't need
+        // a cache entry at all and the dictionary stays small for the common case.
+        if (!tag.WriteOnChange && !tag.WriteDeadband.HasValue) return;
+
+        _lastValues[(deviceHostAddress, tag.TagPath)] = writtenValue;
+    }
+
+    /// <summary>
+    ///     Drop every cached last-value for one device. Called on reconnect or driver shutdown
+    ///     so the next write after a wire-state change pays the full round-trip — the PLC may
+    ///     have been restarted and our cached "we already wrote 42" is stale.
+    /// </summary>
+    public void Reset(string deviceHostAddress)
+    {
+        ArgumentNullException.ThrowIfNull(deviceHostAddress);
+
+        // ConcurrentDictionary doesn't have a "remove where" overload, so iterate keys + remove.
+        // Suppression races are tolerated — losing one suppression decision after a reconnect
+        // costs at most one extra wire write, never correctness.
+        foreach (var key in _lastValues.Keys)
+        {
+            if (string.Equals(key.Device, deviceHostAddress, StringComparison.OrdinalIgnoreCase))
+                _lastValues.TryRemove(key, out _);
+        }
+    }
+
+    /// <summary>Drop every cached last-value across all devices — invoked on full driver shutdown.</summary>
+    public void ResetAll() => _lastValues.Clear();
+
+    private static bool TrySuppress(AbCipTagDefinition tag, object? lastValue, object? newValue)
+    {
+        // Numeric deadband — only fires when both sides convert cleanly to double. NaN / Infinity
+        // bypass: the wire decides because IEEE-754 comparisons against NaN are undefined and
+        // we don't want a stale +Inf in the cache to silently swallow a real reset.
+        if (tag.WriteDeadband.HasValue
+            && TryToDouble(lastValue, out var lastNum)
+            && TryToDouble(newValue, out var newNum))
+        {
+            if (double.IsNaN(lastNum) || double.IsNaN(newNum)
+                || double.IsInfinity(lastNum) || double.IsInfinity(newNum))
+            {
+                // Fall through to the WriteOnChange equality check below — NaN / Infinity skip
+                // the deadband path but a legacy WriteOnChange tag should still benefit from
+                // exact-equality suppression on the same packet.
+            }
+            else if (Math.Abs(newNum - lastNum) < tag.WriteDeadband.Value)
+            {
+                return true;
+            }
+        }
+
+        // WriteOnChange — equality fallback. Always evaluated when the flag is set so a
+        // non-numeric tag (BOOL, STRING) still benefits even when WriteDeadband is set on the
+        // same tag (the deadband path simply doesn't apply to it).
+        if (tag.WriteOnChange && Equals(lastValue, newValue))
+            return true;
+
+        return false;
+    }
+
+    private static bool TryToDouble(object? value, out double result)
+    {
+        // IConvertible covers every Logix atomic type the AB CIP driver decodes (sbyte, short,
+        // int, long + their unsigned siblings + float / double). DateTime and string are
+        // excluded — neither has a meaningful "deadband" interpretation.
+        switch (value)
+        {
+            case null:
+                result = 0;
+                return false;
+            case bool:
+                result = 0;
+                return false;
+            case string:
+                result = 0;
+                return false;
+            case DateTime:
+                result = 0;
+                return false;
+            case IConvertible conv:
+                try
+                {
+                    result = conv.ToDouble(System.Globalization.CultureInfo.InvariantCulture);
+                    return true;
+                }
+                catch
+                {
+                    result = 0;
+                    return false;
+                }
+            default:
+                result = 0;
+                return false;
+        }
+    }
+
+    private sealed class LastKeyComparer : IEqualityComparer<(string Device, string Tag)>
+    {
+        public static readonly LastKeyComparer Instance = new();
+
+        public bool Equals((string Device, string Tag) x, (string Device, string Tag) y) =>
+            string.Equals(x.Device, y.Device, StringComparison.OrdinalIgnoreCase)
+            && string.Equals(x.Tag, y.Tag, StringComparison.Ordinal);
+
+        public int GetHashCode((string Device, string Tag) obj) =>
+            HashCode.Combine(
+                StringComparer.OrdinalIgnoreCase.GetHashCode(obj.Device),
+                StringComparer.Ordinal.GetHashCode(obj.Tag));
+    }
+}

tests/ZB.MOM.WW.OtOpcUa.Driver.AbCip.IntegrationTests/AbCipWriteDeadbandTests.cs

@@ -0,0 +1,87 @@
+using Shouldly;
+using Xunit;
+using ZB.MOM.WW.OtOpcUa.Core.Abstractions;
+using ZB.MOM.WW.OtOpcUa.Driver.AbCip;
+
+namespace ZB.MOM.WW.OtOpcUa.Driver.AbCip.IntegrationTests;
+
+/// <summary>
+///     PR abcip-4.2 — end-to-end coverage for write-deadband / write-on-change suppression
+///     against a running <c>ab_server</c>. Drives a 5-write jittery sequence with
+///     <c>WriteDeadband=1.0</c> and asserts the driver's <c>AbCip.WritesSuppressed</c>
+///     diagnostics counter reflects the expected number of suppressions. Wire-level write
+///     count isn't directly observable in <c>ab_server</c> (no admin shim for "tell me how
+///     many CIP writes you got"), so the suppression evidence is the driver's own counter
+///     plus the final read confirming the last passed-through value reached the PLC.
+/// </summary>
+/// <remarks>
+///     Unit coverage in <see cref="ZB.MOM.WW.OtOpcUa.Driver.AbCip.Tests.AbCipWriteDeadbandTests"/>
+///     proves the suppression math against an in-process fake. This test exercises the full
+///     libplctag stack so a regression in how the driver wires its coalescer to the real wire
+///     path shows up here.
+/// </remarks>
+[Trait("Category", "Integration")]
+[Trait("Requires", "AbServer")]
+public sealed class AbCipWriteDeadbandTests
+{
+    [AbServerFact]
+    public async Task Jittery_setpoints_within_deadband_dont_reach_the_wire()
+    {
+        var profile = KnownProfiles.ControlLogix;
+        var fixture = new AbServerFixture(profile);
+        await fixture.InitializeAsync();
+        try
+        {
+            var deviceUri = $"ab://127.0.0.1:{fixture.Port}/1,0";
+            var drv = new AbCipDriver(new AbCipDriverOptions
+            {
+                Devices = [new AbCipDeviceOptions(deviceUri, profile.Family)],
+                // ab_server seeds TestDINT — drive integer setpoints with a 1.0 deadband so
+                // values that differ by 0 are suppressed. Real-world deadbanding usually
+                // targets REAL setpoints; integer here is fine because the suppression rule
+                // looks at the boxed numeric value, not the on-wire encoding.
+                Tags = [new AbCipTagDefinition("Setpoint", deviceUri, "TestDINT",
+                    AbCipDataType.DInt, WriteDeadband: 1.0)],
+                Timeout = TimeSpan.FromSeconds(5),
+            }, "drv-write-deadband-smoke");
+
+            await drv.InitializeAsync("{}", TestContext.Current.CancellationToken);
+
+            // Five-write jittery sequence: 100, 100, 100, 102, 102.
+            // - 100 (first): passes (no prior).
+            // - 100, 100: suppressed (|0| < 1.0).
+            // - 102: passes (|2| ≥ 1.0).
+            // - 102: suppressed (|0| < 1.0).
+            // Expected: 2 wire writes, 3 suppressions.
+            var inputs = new[] { 100, 100, 100, 102, 102 };
+            foreach (var v in inputs)
+            {
+                var results = await drv.WriteAsync(
+                    [new WriteRequest("Setpoint", v)],
+                    TestContext.Current.CancellationToken);
+                results.Single().StatusCode.ShouldBe(AbCipStatusMapper.Good,
+                    "every write — suppressed or not — must surface as Good to the OPC UA client");
+            }
+
+            drv.WriteCoalescer.TotalWritesSuppressed.ShouldBe(3);
+            drv.WriteCoalescer.TotalWritesPassedThrough.ShouldBe(2);
+
+            // Final readback proves the last passed-through value (102) made it to the PLC.
+            var readback = await drv.ReadAsync(["Setpoint"], TestContext.Current.CancellationToken);
+            readback.Single().StatusCode.ShouldBe(AbCipStatusMapper.Good);
+            Convert.ToInt32(readback.Single().Value).ShouldBe(102);
+
+            // Diagnostics counters are also reflected through GetHealth — the path the
+            // driver-diagnostics RPC + Admin UI consume.
+            var diag = drv.GetHealth().DiagnosticsOrEmpty;
+            diag["AbCip.WritesSuppressed"].ShouldBe(3);
+            diag["AbCip.WritesPassedThrough"].ShouldBe(2);
+
+            await drv.ShutdownAsync(TestContext.Current.CancellationToken);
+        }
+        finally
+        {
+            await fixture.DisposeAsync();
+        }
+    }
+}

tests/ZB.MOM.WW.OtOpcUa.Driver.AbCip.Tests/AbCipWriteDeadbandTests.cs

@@ -0,0 +1,281 @@
+using System.Text.Json;
+using Shouldly;
+using Xunit;
+using ZB.MOM.WW.OtOpcUa.Core.Abstractions;
+using ZB.MOM.WW.OtOpcUa.Driver.AbCip;
+
+namespace ZB.MOM.WW.OtOpcUa.Driver.AbCip.Tests;
+
+/// <summary>
+///     PR abcip-4.2 — write-deadband / write-on-change suppression in
+///     <see cref="AbCipDriver.WriteAsync"/>. The driver consults
+///     <see cref="AbCipWriteCoalescer"/> before issuing any wire write; tests assert the
+///     suppression rules + that suppressed writes still return <c>Good</c> + that the
+///     diagnostics counters increment in lockstep.
+/// </summary>
+[Trait("Category", "Unit")]
+public sealed class AbCipWriteDeadbandTests
+{
+    private const string Device = "ab://10.0.0.5/1,0";
+
+    private static (AbCipDriver drv, FakeAbCipTagFactory factory) NewDriver(params AbCipTagDefinition[] tags)
+    {
+        var factory = new FakeAbCipTagFactory();
+        var drv = new AbCipDriver(new AbCipDriverOptions
+        {
+            Devices = [new AbCipDeviceOptions(Device)],
+            Tags = tags,
+        }, "drv-deadband", factory);
+        return (drv, factory);
+    }
+
+    [Fact]
+    public async Task WriteDeadband_suppresses_intermediate_jittery_setpoints()
+    {
+        // 0.5 deadband; sequence 10.0 → 10.3 → 10.4 → 10.6 — only 10.0 (first write, no prior)
+        // and 10.6 (|10.6 - 10.0| = 0.6 ≥ 0.5) hit the wire. 10.3 + 10.4 are within 0.5 of 10.0
+        // and get suppressed.
+        var (drv, factory) = NewDriver(
+            new AbCipTagDefinition("Setpoint", Device, "Setpoint", AbCipDataType.Real,
+                WriteDeadband: 0.5));
+        await drv.InitializeAsync("{}", CancellationToken.None);
+
+        await drv.WriteAsync([new WriteRequest("Setpoint", 10.0)], CancellationToken.None);
+        await drv.WriteAsync([new WriteRequest("Setpoint", 10.3)], CancellationToken.None);
+        await drv.WriteAsync([new WriteRequest("Setpoint", 10.4)], CancellationToken.None);
+        var last = await drv.WriteAsync([new WriteRequest("Setpoint", 10.6)], CancellationToken.None);
+
+        last.Single().StatusCode.ShouldBe(AbCipStatusMapper.Good);
+        factory.Tags["Setpoint"].WriteCount.ShouldBe(2,
+            "WriteDeadband=0.5 must suppress jittery values within the band");
+        factory.Tags["Setpoint"].Value.ShouldBe(10.6);
+
+        drv.WriteCoalescer.TotalWritesSuppressed.ShouldBe(2);
+        drv.WriteCoalescer.TotalWritesPassedThrough.ShouldBe(2);
+    }
+
+    [Fact]
+    public async Task WriteOnChange_suppresses_repeated_identical_values()
+    {
+        var (drv, factory) = NewDriver(
+            new AbCipTagDefinition("Counter", Device, "Counter", AbCipDataType.DInt,
+                WriteOnChange: true));
+        await drv.InitializeAsync("{}", CancellationToken.None);
+
+        await drv.WriteAsync([new WriteRequest("Counter", 5)], CancellationToken.None);
+        await drv.WriteAsync([new WriteRequest("Counter", 5)], CancellationToken.None);
+        var last = await drv.WriteAsync([new WriteRequest("Counter", 5)], CancellationToken.None);
+
+        last.Single().StatusCode.ShouldBe(AbCipStatusMapper.Good);
+        factory.Tags["Counter"].WriteCount.ShouldBe(1, "WriteOnChange must suppress repeated identical writes");
+
+        drv.WriteCoalescer.TotalWritesSuppressed.ShouldBe(2);
+        drv.WriteCoalescer.TotalWritesPassedThrough.ShouldBe(1);
+    }
+
+    [Fact]
+    public async Task WriteDeadband_takes_priority_over_WriteOnChange_for_numerics()
+    {
+        // Numeric tag with both knobs set: deadband is the active rule, so a value just inside
+        // the deadband suppresses even though it is *not* exactly equal. A value exactly equal
+        // also suppresses (deadband path computes |0| < 0.5 = true).
+        var (drv, factory) = NewDriver(
+            new AbCipTagDefinition("Mixed", Device, "Mixed", AbCipDataType.Real,
+                WriteDeadband: 0.5, WriteOnChange: true));
+        await drv.InitializeAsync("{}", CancellationToken.None);
+
+        await drv.WriteAsync([new WriteRequest("Mixed", 100.0)], CancellationToken.None);
+        await drv.WriteAsync([new WriteRequest("Mixed", 100.2)], CancellationToken.None); // within band
+        await drv.WriteAsync([new WriteRequest("Mixed", 100.0)], CancellationToken.None); // exact equal
+
+        factory.Tags["Mixed"].WriteCount.ShouldBe(1);
+        drv.WriteCoalescer.TotalWritesSuppressed.ShouldBe(2);
+    }
+
+    [Fact]
+    public async Task First_write_always_passes_through_when_no_prior_value()
+    {
+        var (drv, factory) = NewDriver(
+            new AbCipTagDefinition("Speed", Device, "Speed", AbCipDataType.DInt,
+                WriteDeadband: 100.0, WriteOnChange: true));
+        await drv.InitializeAsync("{}", CancellationToken.None);
+
+        await drv.WriteAsync([new WriteRequest("Speed", 0)], CancellationToken.None);
+
+        factory.Tags["Speed"].WriteCount.ShouldBe(1, "first write always passes through");
+        factory.Tags["Speed"].Value.ShouldBe(0);
+        drv.WriteCoalescer.TotalWritesSuppressed.ShouldBe(0);
+        drv.WriteCoalescer.TotalWritesPassedThrough.ShouldBe(1);
+    }
+
+    [Fact]
+    public async Task Reset_after_disconnect_lets_same_value_pass_through_again()
+    {
+        var (drv, factory) = NewDriver(
+            new AbCipTagDefinition("Setpoint", Device, "Setpoint", AbCipDataType.Real,
+                WriteOnChange: true));
+        await drv.InitializeAsync("{}", CancellationToken.None);
+
+        await drv.WriteAsync([new WriteRequest("Setpoint", 42.0)], CancellationToken.None);
+        await drv.WriteAsync([new WriteRequest("Setpoint", 42.0)], CancellationToken.None); // suppressed
+
+        factory.Tags["Setpoint"].WriteCount.ShouldBe(1);
+
+        // Simulate reconnect — the PLC may have restarted while we were offline so the cached
+        // "we already wrote 42" is no longer valid PLC state.
+        drv.WriteCoalescer.Reset(Device);
+
+        await drv.WriteAsync([new WriteRequest("Setpoint", 42.0)], CancellationToken.None);
+        factory.Tags["Setpoint"].WriteCount.ShouldBe(2,
+            "post-reset write must pay the full round-trip even when value is unchanged");
+    }
+
+    [Fact]
+    public async Task Two_devices_keep_independent_caches_for_same_tag_address()
+    {
+        const string device2 = "ab://10.0.0.6/1,0";
+        var factory = new FakeAbCipTagFactory();
+        var drv = new AbCipDriver(new AbCipDriverOptions
+        {
+            Devices = [new AbCipDeviceOptions(Device), new AbCipDeviceOptions(device2)],
+            Tags =
+            [
+                new AbCipTagDefinition("DevA", Device, "Pressure", AbCipDataType.Real,
+                    WriteOnChange: true),
+                new AbCipTagDefinition("DevB", device2, "Pressure", AbCipDataType.Real,
+                    WriteOnChange: true),
+            ],
+        }, "drv-multi-device", factory);
+        await drv.InitializeAsync("{}", CancellationToken.None);
+
+        // Write 42.0 to DevA — passes (first), seeds DevA's cache.
+        await drv.WriteAsync([new WriteRequest("DevA", 42.0)], CancellationToken.None);
+        // Write 42.0 to DevB — must also pass (independent cache, no prior value on DevB).
+        await drv.WriteAsync([new WriteRequest("DevB", 42.0)], CancellationToken.None);
+
+        drv.WriteCoalescer.TotalWritesSuppressed.ShouldBe(0,
+            "device-A and device-B must not share a coalescer cache");
+        drv.WriteCoalescer.TotalWritesPassedThrough.ShouldBe(2);
+    }
+
+    [Fact]
+    public async Task Suppressed_write_returns_Good_status()
+    {
+        var (drv, _) = NewDriver(
+            new AbCipTagDefinition("Setpoint", Device, "Setpoint", AbCipDataType.Real,
+                WriteDeadband: 1.0));
+        await drv.InitializeAsync("{}", CancellationToken.None);
+
+        await drv.WriteAsync([new WriteRequest("Setpoint", 50.0)], CancellationToken.None);
+        var suppressed = await drv.WriteAsync([new WriteRequest("Setpoint", 50.5)], CancellationToken.None);
+
+        suppressed.Single().StatusCode.ShouldBe(AbCipStatusMapper.Good,
+            "OPC UA write semantics: a suppressed write must look successful to the client");
+    }
+
+    [Fact]
+    public async Task Diagnostics_counters_surface_through_GetHealth()
+    {
+        var (drv, _) = NewDriver(
+            new AbCipTagDefinition("Counter", Device, "Counter", AbCipDataType.DInt,
+                WriteOnChange: true));
+        await drv.InitializeAsync("{}", CancellationToken.None);
+
+        await drv.WriteAsync([new WriteRequest("Counter", 7)], CancellationToken.None);
+        await drv.WriteAsync([new WriteRequest("Counter", 7)], CancellationToken.None); // suppressed
+        await drv.WriteAsync([new WriteRequest("Counter", 8)], CancellationToken.None);
+
+        var diag = drv.GetHealth().DiagnosticsOrEmpty;
+        diag["AbCip.WritesSuppressed"].ShouldBe(1);
+        diag["AbCip.WritesPassedThrough"].ShouldBe(2);
+    }
+
+    [Fact]
+    public async Task NaN_or_Infinity_bypasses_deadband_suppression()
+    {
+        var (drv, factory) = NewDriver(
+            new AbCipTagDefinition("Sensor", Device, "Sensor", AbCipDataType.Real,
+                WriteDeadband: 1.0));
+        await drv.InitializeAsync("{}", CancellationToken.None);
+
+        // Seed the cache with a NaN — the next write of 100.0 must NOT be suppressed even
+        // though |100 - NaN| comparison is mathematically meaningless. The wire decides.
+        await drv.WriteAsync([new WriteRequest("Sensor", double.NaN)], CancellationToken.None);
+        await drv.WriteAsync([new WriteRequest("Sensor", 100.0)], CancellationToken.None);
+
+        factory.Tags["Sensor"].WriteCount.ShouldBe(2);
+        drv.WriteCoalescer.TotalWritesSuppressed.ShouldBe(0);
+    }
+
+    [Fact]
+    public async Task Tag_without_either_knob_never_consults_coalescer_cache()
+    {
+        // Plain back-compat tag — no WriteDeadband, no WriteOnChange. Three identical writes
+        // all hit the wire; the fast path in ShouldSuppress increments PassedThrough only.
+        var (drv, factory) = NewDriver(
+            new AbCipTagDefinition("Plain", Device, "Plain", AbCipDataType.DInt));
+        await drv.InitializeAsync("{}", CancellationToken.None);
+
+        await drv.WriteAsync([new WriteRequest("Plain", 1)], CancellationToken.None);
+        await drv.WriteAsync([new WriteRequest("Plain", 1)], CancellationToken.None);
+        await drv.WriteAsync([new WriteRequest("Plain", 1)], CancellationToken.None);
+
+        factory.Tags["Plain"].WriteCount.ShouldBe(3);
+        drv.WriteCoalescer.TotalWritesSuppressed.ShouldBe(0);
+        drv.WriteCoalescer.TotalWritesPassedThrough.ShouldBe(3);
+    }
+
+    [Fact]
+    public async Task Dto_round_trip_preserves_WriteDeadband_and_WriteOnChange()
+    {
+        // Ensure the DTO surface mirrors AbCipTagDefinition so config JSON drives the knobs.
+        // Going through the static factory entry point guarantees the field names + casing
+        // match what operators put in their driver-config JSON.
+        var json = """
+        {
+          "Devices": [{ "HostAddress": "ab://10.0.0.5/1,0" }],
+          "Tags": [{
+            "Name": "Setpoint",
+            "DeviceHostAddress": "ab://10.0.0.5/1,0",
+            "TagPath": "Setpoint",
+            "DataType": "Real",
+            "WriteDeadband": 0.25,
+            "WriteOnChange": true
+          }]
+        }
+        """;
+
+        // Build the driver directly through the internal factory entry point so we can swap
+        // in the FakeAbCipTagFactory after construction; the production CreateInstance path
+        // wires a real LibplctagTagFactory which would try to dlopen libplctag at write time.
+        var dto = JsonSerializer.Deserialize<AbCipDriverFactoryExtensions.AbCipDriverConfigDto>(json,
+            new JsonSerializerOptions
+            {
+                PropertyNameCaseInsensitive = true,
+                ReadCommentHandling = JsonCommentHandling.Skip,
+                AllowTrailingCommas = true,
+            })!;
+        // The DTO carries WriteDeadband / WriteOnChange — the round-trip we actually want to
+        // assert is that AbCipTagDto picks them up + AbCipDriverFactoryExtensions.BuildTag
+        // forwards them to AbCipTagDefinition. Re-running the factory entry point would do
+        // that, but a swappable FakeAbCipTagFactory keeps the test fast + offline.
+        var tagDto = dto.Tags!.Single();
+        tagDto.WriteDeadband.ShouldBe(0.25);
+        tagDto.WriteOnChange.ShouldBe(true);
+
+        // Now use the same shape via the static factory + a fake tag factory so the live
+        // driver actually runs the suppression logic + we can confirm the knobs propagated
+        // all the way through to AbCipTagDefinition.
+        var (drv, factory) = NewDriver(
+            new AbCipTagDefinition("Setpoint", Device, "Setpoint", AbCipDataType.Real,
+                WriteDeadband: tagDto.WriteDeadband, WriteOnChange: tagDto.WriteOnChange ?? false));
+        await drv.InitializeAsync("{}", CancellationToken.None);
+
+        await drv.WriteAsync([new WriteRequest("Setpoint", 1.0)], CancellationToken.None);
+        await drv.WriteAsync([new WriteRequest("Setpoint", 1.0)], CancellationToken.None);
+
+        // WriteOnChange round-tripped — second write of identical value was suppressed.
+        factory.Tags["Setpoint"].WriteCount.ShouldBe(1);
+        drv.WriteCoalescer.TotalWritesSuppressed.ShouldBe(1);
+    }
+}